Policy Gradient Methods

Advantage Actor Critic (A2C)

https://huggingface.co/blog/deep-rl-a2c

A3C

From Lecture 3: Policy Gradient and Advantage Estimation from Deep RL Foundation Series, slides here

• So they have two things
• one is updating the value network $\varphi$, one is updating the policy network $\theta$

The update for $\varphi$ is called fitted Value Iteration